Topic: GMO miner B2: 7NM mining within reach? - page 4. (Read 2774 times)

Never wrote that taserz wrote it.

and  a hard wire 2 outlet circuit at 20 amps  would be asked to do 20x 240 = 4800  x 80 percent that is 3840  watts

or worse  20 x 220 = 4400 x 80 percent that is  3520 watts

since this says it is 1950 or 1990    you are doing 3900 watts to  3980 watts

I have gone past 80 percent derate   and in a hotter setup  it has issues.

I would  not want 100 of these  doing  398,000 watts   which could be  3980/4400 =  as high as 90.45 %    vs 80%    Yeah in canada or iceland not that hot

but plenty of hot mining spots are around

if you have robust volts (240)  you are at 3980/4800 =  82.92%  which may be okay

My area is 239 to 243 volts which is helpful.

Amen to that

BM1385 in the S7 is a 28nm chip, and has a bottom-clock efficiency around 0.18J/GH with decent hashrates still around 0.22J/GH; the 700MHz operating point setpoint of 45-chip S7 was past the knee of the efficiency curve and we also lost ~10% of power, board-level, to the 45A single-phase main regulator plus around 50W machine-level just for fans. Chip-level, the BM1385 was not much worse than some others' early 14/16nm designs.

S3 478GH, 28nm silicon, 366W = 1.306GH/W at the chip = 0.766 W/GH
S4 2000GH, 28nm silicon, 1450W =  1.279GH/W "at the wall" (maybe) ~= 0.725 W/Gh
S5 1155GH, 28nm silicon, 590W = 1.958GH/W = 0.511 W/GH (just goes to show what can be down within a chip 'size')
S7 4500GH, 16nm silicon, 1293W = 3.480GH/W at the chip = 0.287 W/GH - a 1.78X factor in power consumption improvement for a 1.75X decrease in die size - pretty linear.
S9 13500GH, 16nm silicon, 1323W = 10.20GH/W at the chip = 0.098 W/GH - a 2.93X improvement in power consumption via better chip design vs. the S7.

So I'm guessing a shinkage of die size can be estimated to make linear improvement in power consumption, perhaps.

Here is hoping this doesn't get deleted for being "off topic" in a speculation forum...

Frankly, one does have to wonder where folks get their 'increase in efficiency' as node size get smaller figures from. Please explain your reasoning. Is it that you expext x amount of size reduction to equal the same x amount of power savings?

Bzzzt. The Real World doesn't work that way.

Lets see, Inno's A1 28nm chip back in late 2013: promoted to run 1.25x faster and 15% lower power than the Real World proved but - when put into a good design the chips still were/are rock-solid. ref the various Dragon miners from 2014 that are still around. I still have my uber-rare 1THs A1 miner from AMT, one of maybe 5-6 ever delivered by AMT before it went under as one of the casualties of the ill-fated, horrifically bad Bitmine.ch A1 hash board designs.

Same year, 2014, had several failed 28nm designs, BFL Monarch, the Minion chip, and others. Most failed because of the extreme power-density of the designs and insufficient cooling systems.

Then BM nailed the 28nm node with their s7 chip. Time passes and the 16nm s9 chips arrived ruling as most efficient (reference needed: last S7 chips vs first S9 chips. Did the size drop match eff increase? No, but how far off?) but even as early as the s4 BM began pushing power density. Their S5+ began their current shoe box packaging now copied by everyone, warts and all.. Everyone but Canaan with their Avalons that is. Only they seem to have taken the time and thought out proper internal design of the hash boards and heat sinks....

Back to node size/efficiency: The 10nm eBang eBit and Halong/Inno Dragonmint T1 miners began shipping early this year. The eBang 10nm efficiency is worse than their earlier 16nm miners plus it seems the their miners are dying on the vine. The Halong 10nm efficiency cannot really be compared due to the (minor) effective hash rate increase from it using asicboost. That said, despite -ck doing extensive performance tweaking (paid by MyRig and NOT Halong btw) people are reporting the T1 efficiency at advertised speed is less than a s9. Only when dropped to top s9 speeds of around 14THs and less does it match/slightly beat the s9.

Now GMO is pre-selling their 7nm miner with delivery starting Oct (or later). Pre-selling- as did Halong - with advertised specs that said pre-sale Investomers are expecting to achieve with no 'oh by the way the pool must support AB' surprises like Halong pulled or other such crap. Even more power density cooking a node size that even more than the 10nm node is meant for far cooler running, low power (mobile) chips.... Leakage from electrons quantum tunneling between the conductors will be even higher. Hmm, wonder how this will turn out....

But many mid-size farm, like mine, have pre-wired expecting 20 amp 220V to be split between 2 devices, not one. 

I mean you can't really reverse engineer a 7nm chip. Also they didn't bribe anyone to make it faster than the sia they just did it a really inefficient way which allowed them to come to market faster even though the efficiency was shit compared to the others but they didn't care because they flooded the market.

---

I mean the biggest issue I see with a lot of mines is they are all going for high density well at least the services and business I use. So okay yeah you need more cooling to cool it down but if you have the hot and cold isles built for this kind of air movement your right as rain. Also for the 2kw these draws you figure on a 60amp pdu @ 240 you can run 7 of these off a 60amp pdu well maybe wil the rule of 80 your looking at 6 but that is almost using a full load. Granted the bigger miners with big power rails none of this really matter to them as pdu's are not an issue. When I had the majority of miners in my basement my issue was space. I had 40 miners and I did not have a way to put them all into the hot box. The hot box had proper cfm to vent it out but venting in 40 4inch ducts is just a mess when if I can do just 10 ducts I am right as rain. But I see your complaint but those of us and those who have to deal with high density due to limited room these could be great.

Well  1950 watts or 1990 watts either way   it will be roasting hot.

Even if they offer a semi speed control  like -ck did with the T1 (nice work on his part)
Or voltage offset like the Avalon 841

I can state it is a moron move to run 1950-1990 watts on that size and shape.

The density is not that important.
Cooling and efficiency are important.

They have began with a good chip (maybe)  and put it in a moron shape.

How about same size box doing 12th at 972 watts?  with  low medium fast speed settings.

The gear they built  with not like air cooling it will want ac and a colder room  which makes the power savings lost.

All these smart guys  and they sell this in a stupid box.

I tell you what  does this gear come on Nov 3 since it shipped on Oct 31?

And if it does come on Nov 3   How much do you want to bet  that bitmain beats them to market with the s-11   with  17 th doing 1377 watts.  same efficiency and cost of 1400?

How will GMO look then.

They really fucked up doing a pre order  with Oct delivery.

So what, they're using air with a higher specific heat? Epoxy with zero thermal resistance? 300CFM fans that don't spin at 10KRPM and draw 100W to do it? Have they discovered a heatsink profile with no solidity and perfect heat transfer, and made sure to place every single one to avoid vortices?

Quality or not, the physics of air cooling is a big limitation. Look at all the issues we already have with heat unreliability on 16nm and 10nm and think how much worse it's going to be on 7nm, which has basically never been tested before this point.

GMO designed the machine in Japan, which is known for high quality. This means their rigs may outperform the china competitors in terms of quality.

Designed in Japan, produced in Taiwan



I would not be surprised if they offer 1 year warranty. This can be a big advantage.

Yes can do, but you'll void the warranty on the S9s, You can do that on the 841s already but so far I found it unnecessary since -2 turns good yield with good power draw already.


I think GMO maybe releasing a smaller miner for home users late this year or early next year thats got lower power draw and lower speed from the looks of it. Apparently this machine is for "Large server farms". Also if you haven't noticed, the reason the price is so high is because they are mining themselves and they're not going to sell it at lower premiums

yeah  over clocking morons  caused lots of returns.

Sooo   no more voltage  controls.

To me I rather run all gear   at 60 to 80 percent of tops speeds and volts.

But no can do.

I've also been saying that full control of core clock and voltage are necessary features but nobody does that anymore. Probably for idiot-proofing. It is handy some have at least partial options.

add on the lack of speed control.

they will run balls to the wall.

Avalon has various settings one is about 990 watts

Halong t1 with -ck firmware low setting is about 1300 watts

I've been saying exactly that for three years, since S7 rolled out, but it looks like that kind of high-dense chipsinked package is becoming everyone's de-facto standard (except Avalon, thankfully).

There's no good reason not to build KW/sub-KW miners. Site density stopped being an issue back when people started using shelves instead of racks.

1.4KW is already difficult in that case size, without adding half again to it. Especially since all evidence points to smaller nodes having more issues with heat density in the dies, so the chips probably can't be allowed to run continuously at 120C die temperatures without a high risk of roasting before the machine's expected end of viability.

Okay so let us say that GMO is bigger $$$

they still went in the wrong direction here.

Most miners even 1000 unit miners have space to spare
They are short with cheap power
They are short with cooling issues.

I can run 3x my current amount of miners and still have space left over
I have lots of space.
Packing all the power into a small box is dumb and asks for trouble.

Forcing a use of a 2200 watt psu and pulling 1990 watts against it is meh.

Running a 12th doing 995 watts is so easy to manage.

Well Since I can't get these until October it is not a worry for me.

GMO is using TSMC as well also don't forget GMO coming in has a lot more money than all these other players. Even bitmain so it's not like they are coming to the game short handed.

I wanted to talk about that a little bit. They have 4 PCI connectors per board because of the power-draw. I do not think these will work at any place +/- 40° from the equator. At least not without serious cooling problems. From what my friend told me about the presentation, they can be daisy-chained like the avalons so you won't need alot of splitters.

Also, 1990W is apparently the figure AT THE WALL with their PSU. I think that has already been clarified. 2k is already including the PSU. Unfortunately for everyone, MOQ is 100 units. Also..........apparently they had alot of help from TSMC, which in proxy means lol, they ripped off bitmain alot.

Probably because it's still speculation that they have more than just a prototype. They could have run it at the presentation to show it was legit. On top of that, their photos show a hash board that has a completely different layout and less than half the chips that they show in renderings here:
https://www.youtube.com/watch?v=6TcVtEXbjik

this product  is designed in a stupid manner.

a 1950 watt miner  in the same shape and size as the s9 or t1 is pretty much an exercise in stupidity.

Owning the s9
Owning the avalon 841
Owning the t1

They all are close to 1300-1500 watts

trying to cool 1950 watts  in about the same space and size is going to be harder.

also if you have 30 amp pdus  3x 1950 = 5850 watts which will over load the pdu

but if it does 0.08 watts it is good for a guy with limited power.

Ie a smaller miner  maybe a guy with a 30 amp pdu   as he can have 2 of them and 1 s9

1950
1950
1350
5250  watts is a good load for a 30 amp pdu

gives him 24 + 24 + 14 = 62th

while 4x s9i = 4 x 14th = 56th
using 4x 1350 = 5400 watts

I know of a few guys that prepaid  data centers  for 10kwatts a month for 12
 months.

5 of these = 9750 watts and do 5 x 24 = 120th

7  s9i =  9450 watts and do 7 x 14 = 98th

I am not sure that  22th  over the space of a year will earn  the difference  you paid extra.

I also think these use 13  pcie jacks.  so 1950/12 = 162.50 watts a hashboard jack

the s9i is 1350/9  150 per jack

see photo



My guess is  this is not going to be so good

They were better off  making it do 14.00 th  and 1100 watts